change trusting claude

demos/melodyflow_app.py

@@ -1,7 +1,7 @@
 # Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
 
-# This source code is licensed under thmage license found in the
+# This source code is licensed under the license found in the
 # LICENSE file in the root directory of this source tree.
 
 import spaces
@@ -15,6 +15,7 @@ from tempfile import NamedTemporaryFile
 import time
 import typing as tp
 import warnings
+import base64
 
 import torch
 import gradio as gr
@@ -201,11 +202,18 @@ def predict(model, text,
         duration=duration,
         progress=True,)
 
-    # Read the audio file and return as numpy array
+    # Read the audio file and convert to base64
     wav_path = wavs[0]
-    audio_array, sample_rate = audio_read(wav_path)
+    with open(wav_path, 'rb') as f:
+        audio_bytes = f.read()
     
-    return (sample_rate, audio_array.squeeze().cpu().numpy())
+    audio_base64 = base64.b64encode(audio_bytes).decode('utf-8')
+    
+    # Return as a dictionary with base64 data
+    return {
+        "audio": audio_base64,
+        "format": "wav"
+    }
 
 
 def toggle_audio_src(choice):
@@ -272,7 +280,7 @@ def ui_local(launch_kwargs):
                         label="Regularization Strength", minimum=0.0, maximum=1.0, value=0.0, interactive=False)
             with gr.Column():
                 audio_outputs = [
-                    gr.Audio(label=f"Generated Audio - variation {i+1}", type='numpy', show_download_button=False, show_share_button=False) for i in range(N_REPEATS)]
+                    gr.JSON(label=f"Generated Audio - variation {i+1}") for i in range(N_REPEATS)]
         submit.click(fn=predict,
                      inputs=[model, text,
                              solver,
@@ -391,7 +399,7 @@ def ui_hf(launch_kwargs):
                         label="Regularization Strength", minimum=0.0, maximum=1.0, value=0.0, interactive=False)
             with gr.Column():
                 audio_outputs = [
-                    gr.Audio(label=f"Generated Audio - variation {i+1}", type='numpy', show_download_button=False, show_share_button=False) for i in range(N_REPEATS)]
+                    gr.JSON(label=f"Generated Audio - variation {i+1}") for i in range(N_REPEATS)]
         submit.click(fn=predict,
                      inputs=[model, text,
                              solver,